Best of LinkedIn: Health Tech CW 36/ 37

00:00:00: This episode is provided by Thomas Allgeier and Frennis, based on the most relevant LinkedIn posts about health tech from CW-ThirtysixandThirtySeven.

00:00:09: Frenis equips HealthTech providers with a market intelligence to identify which hospitals to target.

00:00:24: So have you noticed that walking through a health tech convention lately feels exactly like I don't know wandering through an auto show?

00:00:32: Yeah, completely.

00:00:33: You see the the holograms The surgical robots the AI that promises to basically cure aging.

00:00:39: right

00:00:39: it's a dazzling.

00:00:40: But you kind of walk away thinking you would never actually trust any Of those concept cars on the actual highway

00:00:44: because for a long time you really couldn't.

00:00:46: i mean It was all pilots.

00:00:47: we were operating in these highly controlled Sandbox environments Because healthcare systems By design, they resist that whole Silicon Valley ethos of moving fast and breaking things.

00:00:57: You just can't bait a test with patient

00:00:59: safety.".

00:01:00: No definitely not.

00:01:01: but looking at the market intelligence you're navigating today for our deep dive into the top health tech trends across LinkedIn That dynamic has completely inverted

00:01:11: it really has.

00:01:12: The concept cars are suddenly parked in the driveway you know, speculating about is actively becoming the embedded invisible infrastructure of hospitals.

00:01:22: Right we are seeing a really hard transition from that pilot phase of innovation directly into actual clinical reality.

00:01:30: it's

00:01:30: So for you listening, we're making this our mission today.

00:01:33: We aren't just looking at isolated gadgets—we are going to unpack a fundamental rewiring of health tech!

00:01:39: We've got four main clusters from the source material to get

00:01:42: through.

00:01:42: Yeah — we're hitting AI regulation and clinical validation first then AI embedded in clinical workflows moving into diagnostics and imaging and wrapping up with care delivery and market strategy.

00:01:52: Exactly – And the very foundation of this shift.

00:01:55: it starts with a bottleneck that has frustrated developers.

00:02:00: How do we safely approve the absolute flood of AI hitting the market right now?

00:02:04: A huge question.

00:02:05: Just to give you a sense of scale here, Joel Kent pointed out that back in twenty nineteen The FDA authorizing eighty AI enabled medical devices felt like a major industry milestone.

00:02:16: Oh yes!

00:02:17: Huge news back then

00:02:18: But NOW We are pacing toward three hundred and sixty two authorizations In twenty twenty six alone.

00:02:26: Wow I mean, that is essentially an AI medical device cleared every single day.

00:02:31: It really has shifted from this rare novelty to just routine administrative processing.

00:02:37: but the underlying mechanics of how regulators are keeping pace without you know compromising patient safety That's what truly brilliant here.

00:02:45: Right

00:02:46: because traditional software regulation was built for static code.

00:02:48: Exactly You engineer it and run through clinical trials The FDA approves Basically never changes.

00:02:53: If you want to change it, You start over.

00:02:55: but AI is fundamentally designed To learn to adapt and shift based on new data

00:02:59: Which totally shatters the old regulatory paradigm?

00:03:02: You can't just Resubmit to the FDA every time a machine learning model gets slightly better at its job.

00:03:07: No, you'd be filing paperwork Every hour.

00:03:10: And this Is where The concept of predetermined Change control Plans Comes into play Or PCCPs.

00:03:16: A Syncon recently analyzed This framework.

00:03:19: It's fascinating.

00:03:20: A PCCP is essentially a binding pre-negotiated contract with the FDA.

00:03:26: Okay, so it's not a blank check for your algorithm to just learn wild new tricks on the open internet?

00:03:31: Definitely not!

00:03:32: Instead... ...a developer defines an exact envelope of acceptable change

00:03:37: Like mapping out future boundaries up front.

00:03:39: Right

00:03:40: You tell the regulator Our model might shift its weighting as it sees more diverse patient data but will never operate outside these specific pretested parameters.

00:03:49: So you establish what continuous validation looks like before the software ever even touches a hospital server.

00:03:54: Precisely, and this isn't just an isolated FDA experiment either.

00:03:58: Christian Hein observed really interesting global convergence happening right now.

00:04:02: Oh

00:04:02: Really?

00:04:03: Between who?

00:04:04: between Europe's EMA and USFDA.

00:04:06: They are aligning on performance based approach especially for generative AI.

00:04:11: Okay so What does that look in practice?

00:04:13: Well, rather than solely auditing the opaque mathematics of an underlying foundation model they are pivoting to evaluate these systems based on demonstrated clinical competencies.

00:04:24: That makes a lot sense.

00:04:25: actually you don't credential human surgeon by dissecting their brain chemistry.

00:04:30: no you credential them based on their demonstrated ability perform surgery safely.

00:04:35: we're starting treat autonomous system in same way

00:04:38: exactly And that framework is opening the door for tools that actively share The Clinical Load.

00:04:44: Anders Ponce shared a perfect example with VA Era.

00:04:47: Yeah, they just secured CE certification in Europe For An Autonomous AI Breast Cancer Screening System.

00:04:54: and this isn't Just A Spell Checker for Radiologists.

00:04:57: it analyzes mammography images autonomously which dramatically reduces the necessity for multiple human radiologists to double check every single baseline healthy case.

00:05:10: Yeah,

00:05:21: it communicates with the patient tracks their vitals and alters the dosage within guardrails preset by the physician.

00:05:28: That directly addresses the reality that you know.

00:05:31: chronic illness doesn't politely pause between quarterly thirty minute checkups

00:05:37: right?

00:05:37: The AI manages the micro adjustments in real time.

00:05:40: but hold on I have to push back a bit here.

00:05:43: If we are adopting this progress over perfection mindset to clear these autonomous tools faster, aren't we just rushing past the guardrail?

00:05:51: It's a valid concern.

00:05:52: Because

00:05:53: Sharna Satiraj you raised flags about the new integration between open AI and EPIC—the largest electronic health record system.

00:06:00: Oh yeah!

00:06:00: The hallucination risk?

00:06:01: Exactly We're wiring generative AI which know can hallucinate directly into central nervous systems of patient data.

00:06:08: The argument is that we need to act fast, to save lives and reduce clinician burnout.

00:06:13: But man the stakes of an LLM making a charting error are catastrophic.

00:06:17: It IS the ultimate tension in health tech right now.

00:06:21: You can't let perfect be the enemy of good but reckless.

00:06:24: deployment at scale Is incredibly dangerous And Elena Schwetz pointed out A very specific mechanical blind spot.

00:06:32: That happens when you move too fast Even in heavily regulated spaces like clinical trials.

00:06:37: What kind of blind spot?

00:06:38: Well, using AI to screen patient populations for a trial is blazingly fast.

00:06:44: We are talking about algorithms cutting a forty-five hour human review process down two and half hours.

00:06:48: I mean that sounds like massive win for trial efficiency.

00:06:51: It IS until you look at how the algorithm makes its shortlist.

00:06:55: An AI model Only knows the universe of data it was trained on.

00:06:59: Oh,

00:06:59: I see where this is going.

00:07:00: Yeah if older adults rural populations or minority groups are systematically underrepresented in historical trial data The algorithm doesn't even know they exist.

00:07:09: so just skips them entirely right?

00:07:11: It will scan a hospital's database and cleanly efficiently bypass those patients.

00:07:16: A faster short list can easily become an exclusionary one If human oversight just assumes the machine as perfectly objective

00:07:22: Which brings up exactly what happens when these cleared technologies actually hit the hospital floor.

00:07:27: because an algorithm can have FDA clearance and zero-algorithmic bias, but if it adds say three extra clicks to a doctor's interface.

00:07:36: It will be violently rejected by the staff

00:07:38: as it should be.

00:07:39: clinicians are drowning in software burdens.

00:07:42: The only AI that survives the clinical floor is AI that actively removes administrative weight.

00:07:48: Do we have hard numbers on that?

00:07:50: Actually happening

00:07:50: yet We do.

00:07:52: Seema Verma and Jeff Hibbert shared a concrete ROI metric that proves this is finally working.

00:07:58: Oracle Health deployed their clinical AI agent at Atlantic Air, And the hard data showed it cut physician documentation time by thirty-seven percent.

00:08:06: Thirty seven percent is staggering!

00:08:08: For context for you listening we are talking about eliminating industry calls.

00:08:11: pajama time

00:08:12: Yes The worst Time.

00:08:13: Doctors sitting in head at midnight typing up the clinical notes from there.

00:08:16: two p m appointments

00:08:17: Exactly.

00:08:18: We're seeing friction removed on patient side too.

00:08:21: Sean Biden noted how Sutter Health deployed an AI assistant named Emmy.

00:08:25: But what's crucial here is that Emmy isn't some generic chatbot on a public website, it lives securely behind the firewall right inside the patient's MyChart portal.

00:08:35: It handles hyper-specific secure questions.

00:08:38: twenty four seven

00:08:40: When you step back, it really feels like AI is maturing into the fundamental operating layer of the hospital.

00:08:46: It's shifting from taking a static snapshot of a patient health to analyzing the entire movie.

00:08:51: I liked

00:08:51: that analogy.

00:08:52: Yeah Ryan Fukushima described the development on oncology foundation models or OFMs.

00:08:58: Standard models look at one tumor biopsy slide or a single genomic profile frozen in time, right?

00:09:04: A snapshot.

00:09:05: but cancer mutates and OFM tracks the chronological timeline analyzing how the disease shifts develops resistance And responds to distinct therapies over months

00:09:14: because of tumors moving target.

00:09:16: if your software only looks at the past you can't predict mutation

00:09:19: Exactly, and Jean-Marie Coferugia discussed how Mayo Clinic is utilizing an AI foundation model to detect hypertrophic cardiomyopathy from routine echocardiograms.

00:09:29: Oh so they aren't even ordering new tests for this?

00:09:32: No!

00:09:32: The AI isn't ordering complex new tests.

00:09:35: it's analyzing the pixel data of routine imaging that are already performing And spotting microscopic patterns that human eyes simply cannot process.

00:09:43: That...is

00:09:43: incredible.

00:09:44: But here is the reality check for anyone building these systems.

00:09:47: None of this software magic functions in a vacuum.

00:09:50: Robin Goldsmith zeroed-in on The massive infrastructure bottleneck we are hitting.

00:09:55: you

00:09:55: mean like network bandwidth exactly?

00:09:58: Think about those ambient AI tools that listen to a patient encounter and write the clinical note in real time.

00:10:03: You cannot run multiple concurrent streams of continuous audio processing through a hospital's legacy Wi-Fi network,

00:10:10: right?

00:10:10: Because if the network drops for even three seconds The AI completely misses the dosage instruction

00:10:16: precisely.

00:10:16: And you cannot dump that continuous optimization burden onto a Hospital IT team That is already struggling to keep the basic servers running.

00:10:24: So what's the fix?

00:10:25: deploying this level of tech requires ultra-secure, managed network services like those provided by Verizon running silently in the background.

00:10:34: Ensuring latency is zero so that clinical tools actually function without crashing the emergency room's

00:10:39: bandwidth."

00:10:40: Which logically forces a massive platform consolidation on the market?

00:10:44: Joshua Liu laid out this strategic roadmap for it... Midsized non-academic health systems, they don't want to manage fifty different AI vendors.

00:10:52: No it's a nightmare!

00:10:53: They are going to standardize on single unified platform whether that is driven by Epic and Enterprise OpenAI deployment or Doximity...they're buying simplicity

00:11:03: And the massive academic centers

00:11:05: The Tier One Academic Medical Centers have IT muscle to prefer hybrid multi vendor approach.

00:11:11: They'll cherry pick absolute best in breed algorithm for each specific disease state

00:11:16: classic tech cycle.

00:11:17: We're moving from the fragmentation of thousands-of point solutions into

00:11:24: the platform.

00:11:33: era

00:11:43: Foundation models are incredibly powerful, but they're entirely dependent on the physical data.

00:11:49: They ingest

00:11:50: right garbage in garbage out.

00:11:51: exactly if a hospital's imaging hardware is outdated and capturing blurry scans The most advanced AI in the world Is effectively blind.

00:12:01: Which brings us to the third theme?

00:12:03: The massive physical transformation happening in diagnostics Imaging and robotics the heavy machinery as being entirely re-engineered.

00:12:12: Dr.

00:12:12: Flupo Katamartiri and Lubislav Yakhman detailed a generational leap in imaging called spectral photon counting CT.

00:12:19: Wait,

00:12:19: let's break down the mechanics of that because it sounds incredibly dense.

00:12:22: Okay, yeah.

00:12:23: Think of a traditional CT scan like a standard black and white photograph.

00:12:27: It measures the total amount of x-ray energy passing through the body Which gives you a great picture of morphology You know?

00:12:34: The general shape and density of organs or tumors.

00:12:37: okay photon counting Ct is like suddenly seeing the body in high definition color.

00:12:42: How does it actually do that

00:12:43: physically?

00:12:44: Instead of measuring total energy, the new detectors literally count individual x-ray photons as they hit and measure a specific energy level for every single photon.

00:12:54: That is wild!

00:12:55: It allows radiologists to perform multi parametric tissue characterization.

00:12:59: They can digitally separate different materials like iodine calcium or specific soft tissues with microscopic

00:13:06: precision.

00:13:06: So that's exactly what you're looking at.

00:13:09: It is completely redefining diagnostics for complex spaces like pelvic masses, and a lot of the critical deep device testing for this is being driven by specialized teams in Slovakia right now.

00:13:24: And Anjay Helwitsch highlighted a similar mechanical leap in MRI technology with radial UTE-MRI.

00:13:30: Okay,

00:13:30: what is that?

00:13:31: Well conventional MRI creates images based on how long it takes hydrogen protons and the body to release energy A process called relaxation time.

00:13:39: Right

00:13:40: But certain tissues like cortical bone or highly calcified cartilage have What's called a short T two relaxation Time.

00:13:47: they released their energy so incredibly fast That The standard MRI scanner just misses the signal entirely.

00:13:53: Oh, so in a normal MRI bone shows up as dark invisible void?

00:13:58: Exactly!

00:13:59: Radial UTE stands for Ultra-Short Echo Time.

00:14:02: The hardware and software have been redesigned to capture that fleeting signal before it fades giving clinicians almost CT level contrasts and detail for bone structures but without blasting patient with ionizing radiation.

00:14:13: Wow And beyond physics of imaging itself Hardware innovation is actually altering physical layout A tool.

00:14:20: Gupta noted the deployment of a new helium-free interventional MRI suite developed by Phillips and Imrecor.

00:14:27: Okay, designed for electrophysiology right?

00:14:29: But why does removing helium actually matter?

00:14:32: Because traditional MRI machines require thousands of liters of liquid helium to keep their magnets super cold.

00:14:39: That means you need massive ventilation pipes in specialized reinforced rooms.

00:14:43: So it's an infrastructure nightmare.

00:14:44: Exactly A helium-free MRI seals a tiny amount of coolant permanently inside the machine.

00:14:51: Suddenly, you can install powerful MRI machines directly in surgical suite where it previously couldn't fit giving surgeons real time imaging during complex cardiac procedures.

00:15:01: But

00:15:02: doesn't this create new bottleneck?

00:15:04: I mean, we're giving surgeons photon counting CTs short T two MRI's live interventional imaging.

00:15:10: Are we actually making the physical procedure easier for this surgeon or are we just overwhelming them with an impossible amount of data and screens to interpret while they literally have a scalpel in their hand?

00:15:19: That is the exact friction point engineering teams are trying to solve right now.

00:15:23: And The solution isn't more screens it seamless physical integration.

00:15:27: Martin Zaff shared open access paper that illustrates this beautifully.

00:15:32: They used digital twins to optimize something as purely mechanical,

00:15:39: Wait, a digital twin for a table?

00:15:41: Yes.

00:15:41: Because when a heavy patient is on a cantilevered table sliding into an MRI bore the table sags by microscopic amounts.

00:15:49: Oh and that mechanical sag blurs the high-resolution image

00:15:52: Exactly So they built a heavy physics model to predicts the exact mechanical wear & bend of hardware And paired it with lightweight digital twins that adapts to movement in real time.

00:16:02: The software anticipates the physical sag and corrects image capture automatically.

00:16:15: That's amazing.

00:16:16: And if you want the ultimate proof of software imaging and hardware merging into one seamless tool, just look at the absolute explosion in surgical robotics.

00:16:25: Medtronic is moving aggressively into this space.

00:16:28: Chris Isso noted they have executed fifteen deals.

00:16:34: That's a massive

00:16:35: push right including a major investment in cornerstone robotics and Matt Anderson pointed out the strategy behind that Metronic is positioning.

00:16:43: The cornerstones entire system alongside their existing Hugo robotic system in select international markets.

00:16:49: Yeah,

00:16:50: so they realize health systems don't want one size fits all robots

00:16:53: exactly.

00:16:53: They wanna fleet of specific tools.

00:16:55: they can scale across different surgical departments

00:16:57: And we are seeing the results of this integration live.

00:17:00: Paveena Jiravizakul shared a case study from Advent Health Avista in Colorado, they're utilizing the Stealth XIS autopilot for robotic assisted spine surgery.

00:17:10: It is no longer just mechanical arm.

00:17:12: it natively integrates pre-op imaging The real time spatial navigation and the robotic guidance into one unified workflow for the spine surgeon.

00:17:20: But as look at these incredible advancements autonomous AI to robotic spine surgery, we have to confront a much broader strategic reality.

00:17:29: And this is our fourth theme

00:17:30: right the Reality Check.

00:17:32: yeah does any of these highly engineered technology actually reach the end patient and Does it meaningfully improve care when It gets there?

00:17:41: It's the most important question because health tech doesn't exist in a vacuum.

00:17:45: it gets deployed into highly fragmented, highly flawed human system and Clarita Higgins surfaced a sobering market reality regarding medical device equity.

00:17:54: What did she find?

00:17:55: Hospital systems are increasingly making cost-driven purchasing decisions for basic third party pulse oximetry sensors aiming to shave dollars off their supply chain budgets.

00:18:05: But a pulse oximeter isn't just a generic plastic clip.

00:18:09: It has an optical system that measures light absorption through the skin to determine blood oxygen,

00:18:13: right?

00:18:14: When a hospital alters that system by swapping in cheaper third-party sensors, they introduce massive performance gaps.

00:18:21: And we already have well documented systemic skin tone disparities where these optical devices routinely struggle to accurately read oxygen levels in patients with darker skin masking hidden hypoxia.

00:18:34: That is deeply concerning.

00:18:35: cost driven procurement completely ignores the physics of these disparities.

00:18:40: this strategic question becomes How many delayed interventions or ICU transfers is that fractional cost savings actually worth?

00:18:48: It exposes a massive flaw in how we trust data.

00:18:51: We assume because the sensor spits out.

00:18:53: That number is the absolute truth about a patient's health.

00:18:56: Yeah, and Gels-Reidman highlighted his study on consumer wearables.

00:18:59: that completely shatters this illusion.

00:19:01: We have millions of people wearing advanced smartwatches with sophisticated accelerometers in heart rate sensors.

00:19:06: yet The data shows those variables only explain between two point five to sixteen point.

00:19:10: two percent Of patients self perceived sleep quality

00:19:13: wait Only up to sixteen percent?

00:19:15: yeah...that gap exists because A wearable Is merely measuring kinetic movement And pulse rhythm.

00:19:21: But a human being reporting on how well they slept is executing a complex cognitive appraisal.

00:19:27: They are subconsciously weighing their current stress levels, Their mood there joint pain and their memory of the past week

00:19:34: right?

00:19:34: A firmware update cannot replicate subjective Human judgment.

00:19:37: exactly

00:19:38: that intersection of hard data in human reality creates intense strategic challenges for the future.

00:19:44: Dr.. Martha Bokenfeld discussed the emerging capabilities of predictive brain imaging.

00:19:49: Imagine lying in an MRI scanner that generates a digital twin of your neural pathways, and the software can forecast the onset of Alzheimer's disease ten to fifteen years before you experience single memory lapse.

00:20:00: On clinical level I mean it is miraculous!

00:20:02: It gives patient decade to intervene with lifestyle changes or neuroprotective therapeutics for their options narrow.

00:20:09: But strategically how does market handle this?

00:20:11: Who owns a fifteen-year forecast of your cognitive decline?

00:20:15: That's

00:20:15: this scary thought.

00:20:16: If life insurance or long term care providers access that predictive twin, does the patient become uninsurable?

00:20:24: and ultimately will these predictive insights become an out-of-pocket luxury only accessible to the wealthy widening the care gap?

00:20:31: It perfectly illustrates a critical rule of health tech that Drew Logan summarized Regulatory clearance is entirely different from commercial readiness.

00:20:40: You can have Brilliant FDA cleared science, but if the reimbursement codes don't exist or The sales field fails to explain value proposition To a hospital procurement committee Or tech doesn't integrate into billing workflow.

00:20:54: The product will just fail in market.

00:20:56: You can build perfect science that patients never actually get touch

00:20:59: Which is exactly why Donna Crier argued That industry needs evolve beyond performative patient engagement.

00:21:05: Pulling up patient advisory board on corporate slide deck isn't enough anymore.

00:21:10: We need actual patient leadership integrated into drug development and AI governance, dictating how the technology is built and distributed from day

00:21:17: one.".

00:21:18: And clinicians are echoing that exact sentiment.

00:21:27: Burnt Anasaurage and Burt Van Meers both emphasize that the clinical workforce is fundamentally exhausted.

00:21:33: Completely burned down.

00:21:35: They do not want more complex hyper-engineered technology just for the sake of innovation, they are demanding AI enabled workflow tools that fade into their background to give them time back so they can actually look patients in the eye again.

00:21:48: It

00:21:48: aligns perfectly with Vytran's insights from a recent health tech ecosystem assessment.

00:21:54: True digital transformation doesn't occur by simply air-dropping a massive foundation model into a regional hospital.

00:22:00: No, it never works like that!

00:22:01: It happens through the resilience of the local ecosystem... ...it requires clinicians government regulators and tech developers actively aligning their goals.

00:22:09: Resilient human partnerships dictate success far more than the underlying code itself

00:22:15: which brings us to the core realization this entire shift.

00:22:18: We started by impairing health tech to concept cars at an auto show versus the vehicles you actually trust on the highway.

00:22:26: we now have regulators defining the rules for autonomous algorithms, we have AI agents eliminating a third of clinical documentation time...

00:22:34: ...we have hardware mapping molecular structure of tissues and digital twins guiding robotic surgical arms.

00:22:41: The technology is officially on the Highway!

00:22:43: The

00:22:43: tools work?

00:22:44: The engineering bottleneck has largely been solved

00:22:46: But for YOU THE LISTENER The true implication for anyone building or investing in this space is something Summer Siddiqui pointed out.

00:22:54: Generative AI is exponentially increasing individual leverage.

00:22:58: Today, a single developer writing code can output the work of what used to take an entire engineering team

00:23:03: Building technology itself getting faster easier and cheaper every day Which

00:23:08: means that the code no longer your defensive moat.

00:23:10: Exactly!

00:23:11: When every hospital start-up vendor has tools build concept car the vehicle itself ceases be special.

00:23:17: The real moats in healthcare are shifting, the new motes are trust.

00:23:21: They're distribution channels.

00:23:23: they're a flawless clinical execution.

00:23:25: can you get the procurement committee to buy it?

00:23:27: The exhausted doctor to adopt it without friction and the patient to inherently trust the output?

00:23:33: because moving forward success isn't about having the flashiest algorithm.

00:23:37: It is about having that deepest understanding of human infrastructure.

00:23:40: You were plugging into

00:23:42: That's definitive lens through which view this market.

00:23:45: If you enjoyed this episode, new episodes drop every two weeks.

00:23:48: Also check out our other editions on cloud insights and sovereignty digital products and services AI and agentic systems green ICT in sustainable AI ICT and tech insights and defense tech.